Optimal selection of multicast groups for resource location in a distributed system

Abstract

A protocol is presented to locate (or find) named resources in a distributed system which uses the multicast capabilities of the underlying network. Each node in the network uses a sequence of node groups, and each node group is associated with a unique multicast address. To locate a resource, the searching node sequentially polls each one of the groups until the resource is found; this scheme is a generalization of both pure polling and broadcast. To obtain an optimal division of the nodes into multicast groups, the protocol is analyzed and an efficient algorithm is given that provides a group division minimizing the expected cost per location operation.